def module1(params, train_images):
time_start = time.time()
ph = Pixelhop2(TH1=0.0001,
TH2=0.0001,
SaabArgs=params.SaabArgs,
neighborArgs=params.neighborArgs,
poolingArg=params.poolingArg)
ph.fit(train_images)
print("Time cost - PixelHop++ units:", time.time() - time_start)
del train_images
# save data to files
save_data(ph, os.path.join(params.save_data, "ph"))
'win': 5,
'stride': 1
}, {
'func': ShrinkAvg,
'win': 5,
'stride': 1
}, {
'func': ShrinkAvg,
'win': 5,
'stride': 1
}]
concatArg = {'func': Concat}
p2_avg = Pixelhop2(depth=3,
TH1=0.001,
TH2=0.0001,
SaabArgs=SaabArgs,
shrinkArgs=shrinkArgs,
concatArg=concatArg)
p2_avg.fit(x_train)
# output_train_avg = p2_avg.transform(x_train)
# SaabArgs = [{'num_AC_kernels':-1, 'needBias':False, 'useDC':True, 'batch':None, 'cw':False},
# {'num_AC_kernels':-1, 'needBias':True, 'useDC':True, 'batch':None, 'cw':True},
# {'num_AC_kernels':-1, 'needBias':True, 'useDC':True, 'batch':None, 'cw':True}]
# shrinkArgs = [{'func':ShrinkMax, 'win':7, 'stride':1},
# {'func': ShrinkMax, 'win':5, 'stride':1},
# {'func': ShrinkMax, 'win':3, 'stride':1}]
# concatArg = {'func':Concat}
#
# p2_max = Pixelhop2(depth=3, TH1=0.001, TH2=0.0001, SaabArgs=SaabArgs, shrinkArgs=shrinkArgs, concatArg=concatArg)
# p2_max.fit(x_train)
f.close()
f = open('Train12500.pckl', 'rb')
XTrain12500, YTrain12500 = pickle.load(f)
f.close()
# In[ ]:
#Train module 1 with window size of 3x3
#train pixelHop++ Units
##Module 1
##train 1st unit(depth 1), 2nd unit(depth 2), 3rd unit(depth 3) at once
#constructor arguments are provided in homework instruction
depthOfSSL = 3
pixelHop2Obj = Pixelhop2(depth=depthOfSSL,
TH1=0.001,
TH2=0.0001,
SaabArgs=SaabArgs,
shrinkArgs=shrinkArgs,
concatArg=concatArg)
#training module 1 at all depths
timerObj.tic() #Elapsed time is 246.831189 seconds.
pixelHop2Obj.fit(XTrain6250)
print('Time for training module 1 at all depths: ')
timerObj.toc()
#Save pixelHop2 model
fileName_PixelHop2_6250_3x3 = 'pixelHop2_Cla_6250_3x3.sav'
pickle.dump(pixelHop2Obj, open(fileName_PixelHop2_6250_3x3, 'wb'))
# In[ ]:
'stride': 1,
'num': 5
}]
concatArg = {'func': Concat}
# Training Process
# In[5]:
#Training Time Starts
trainStart = datetime.datetime.now()
#PixelHop Fitting
print("Training the Module 1 of PixelHop")
model = Pixelhop2(depth=5,
TH1=0.0012,
TH2=0.00012,
SaabArgs=SaabArgs,
shrinkArgs=shrinkArgs,
concatArg=concatArg)
model.fit(fitData)
#Using Batching Method To Do the Transform
print("Extracting Features of Training Data")
features_train_raw = PH_Transform(megaSize, batchSize, trainDataSet, model)
#Merge Training Image Features
features_train_layer1, features_train_layer2, features_train_layer3, features_train_layer4, features_train_layer5 = merge_Features(
features_train_raw, megaSize)
#Feature Selection Process
print("Selecting Features of Training Data")
#Calculate the Cross Entrophy for Each Channel
CE1, CE2, CE3, CE4, CE5 = cal_CE(features_train_layer1, features_train_layer2,
features_train_layer3, features_train_layer4,
features_train_layer5, trainLabelSet)
